]>
Commit | Line | Data |
---|---|---|
c0a31329 TG |
1 | /* |
2 | * linux/kernel/hrtimer.c | |
3 | * | |
3c8aa39d | 4 | * Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de> |
79bf2bb3 | 5 | * Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar |
54cdfdb4 | 6 | * Copyright(C) 2006-2007 Timesys Corp., Thomas Gleixner |
c0a31329 TG |
7 | * |
8 | * High-resolution kernel timers | |
9 | * | |
10 | * In contrast to the low-resolution timeout API implemented in | |
11 | * kernel/timer.c, hrtimers provide finer resolution and accuracy | |
12 | * depending on system configuration and capabilities. | |
13 | * | |
14 | * These timers are currently used for: | |
15 | * - itimers | |
16 | * - POSIX timers | |
17 | * - nanosleep | |
18 | * - precise in-kernel timing | |
19 | * | |
20 | * Started by: Thomas Gleixner and Ingo Molnar | |
21 | * | |
22 | * Credits: | |
23 | * based on kernel/timer.c | |
24 | * | |
66188fae TG |
25 | * Help, testing, suggestions, bugfixes, improvements were |
26 | * provided by: | |
27 | * | |
28 | * George Anzinger, Andrew Morton, Steven Rostedt, Roman Zippel | |
29 | * et. al. | |
30 | * | |
c0a31329 TG |
31 | * For licencing details see kernel-base/COPYING |
32 | */ | |
33 | ||
34 | #include <linux/cpu.h> | |
35 | #include <linux/module.h> | |
36 | #include <linux/percpu.h> | |
37 | #include <linux/hrtimer.h> | |
38 | #include <linux/notifier.h> | |
39 | #include <linux/syscalls.h> | |
54cdfdb4 | 40 | #include <linux/kallsyms.h> |
c0a31329 | 41 | #include <linux/interrupt.h> |
79bf2bb3 | 42 | #include <linux/tick.h> |
54cdfdb4 TG |
43 | #include <linux/seq_file.h> |
44 | #include <linux/err.h> | |
237fc6e7 | 45 | #include <linux/debugobjects.h> |
eea08f32 AB |
46 | #include <linux/sched.h> |
47 | #include <linux/timer.h> | |
c0a31329 TG |
48 | |
49 | #include <asm/uaccess.h> | |
50 | ||
c6a2a177 XG |
51 | #include <trace/events/timer.h> |
52 | ||
c0a31329 TG |
53 | /* |
54 | * The timer bases: | |
7978672c GA |
55 | * |
56 | * Note: If we want to add new timer bases, we have to skip the two | |
57 | * clock ids captured by the cpu-timers. We do this by holding empty | |
58 | * entries rather than doing math adjustment of the clock ids. | |
59 | * This ensures that we capture erroneous accesses to these clock ids | |
60 | * rather than moving them into the range of valid clock id's. | |
c0a31329 | 61 | */ |
54cdfdb4 | 62 | DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) = |
c0a31329 | 63 | { |
3c8aa39d TG |
64 | |
65 | .clock_base = | |
c0a31329 | 66 | { |
3c8aa39d TG |
67 | { |
68 | .index = CLOCK_REALTIME, | |
69 | .get_time = &ktime_get_real, | |
54cdfdb4 | 70 | .resolution = KTIME_LOW_RES, |
3c8aa39d TG |
71 | }, |
72 | { | |
73 | .index = CLOCK_MONOTONIC, | |
74 | .get_time = &ktime_get, | |
54cdfdb4 | 75 | .resolution = KTIME_LOW_RES, |
3c8aa39d TG |
76 | }, |
77 | } | |
c0a31329 TG |
78 | }; |
79 | ||
92127c7a TG |
80 | /* |
81 | * Get the coarse grained time at the softirq based on xtime and | |
82 | * wall_to_monotonic. | |
83 | */ | |
3c8aa39d | 84 | static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base) |
92127c7a TG |
85 | { |
86 | ktime_t xtim, tomono; | |
ad28d94a | 87 | struct timespec xts, tom; |
92127c7a TG |
88 | unsigned long seq; |
89 | ||
90 | do { | |
91 | seq = read_seqbegin(&xtime_lock); | |
2c6b47de | 92 | xts = current_kernel_time(); |
ad28d94a | 93 | tom = wall_to_monotonic; |
92127c7a TG |
94 | } while (read_seqretry(&xtime_lock, seq)); |
95 | ||
f4304ab2 | 96 | xtim = timespec_to_ktime(xts); |
ad28d94a | 97 | tomono = timespec_to_ktime(tom); |
3c8aa39d TG |
98 | base->clock_base[CLOCK_REALTIME].softirq_time = xtim; |
99 | base->clock_base[CLOCK_MONOTONIC].softirq_time = | |
100 | ktime_add(xtim, tomono); | |
92127c7a TG |
101 | } |
102 | ||
c0a31329 TG |
103 | /* |
104 | * Functions and macros which are different for UP/SMP systems are kept in a | |
105 | * single place | |
106 | */ | |
107 | #ifdef CONFIG_SMP | |
108 | ||
c0a31329 TG |
109 | /* |
110 | * We are using hashed locking: holding per_cpu(hrtimer_bases)[n].lock | |
111 | * means that all timers which are tied to this base via timer->base are | |
112 | * locked, and the base itself is locked too. | |
113 | * | |
114 | * So __run_timers/migrate_timers can safely modify all timers which could | |
115 | * be found on the lists/queues. | |
116 | * | |
117 | * When the timer's base is locked, and the timer removed from list, it is | |
118 | * possible to set timer->base = NULL and drop the lock: the timer remains | |
119 | * locked. | |
120 | */ | |
3c8aa39d TG |
121 | static |
122 | struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer, | |
123 | unsigned long *flags) | |
c0a31329 | 124 | { |
3c8aa39d | 125 | struct hrtimer_clock_base *base; |
c0a31329 TG |
126 | |
127 | for (;;) { | |
128 | base = timer->base; | |
129 | if (likely(base != NULL)) { | |
3c8aa39d | 130 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
131 | if (likely(base == timer->base)) |
132 | return base; | |
133 | /* The timer has migrated to another CPU: */ | |
3c8aa39d | 134 | spin_unlock_irqrestore(&base->cpu_base->lock, *flags); |
c0a31329 TG |
135 | } |
136 | cpu_relax(); | |
137 | } | |
138 | } | |
139 | ||
6ff7041d TG |
140 | |
141 | /* | |
142 | * Get the preferred target CPU for NOHZ | |
143 | */ | |
144 | static int hrtimer_get_target(int this_cpu, int pinned) | |
145 | { | |
146 | #ifdef CONFIG_NO_HZ | |
147 | if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu)) { | |
148 | int preferred_cpu = get_nohz_load_balancer(); | |
149 | ||
150 | if (preferred_cpu >= 0) | |
151 | return preferred_cpu; | |
152 | } | |
153 | #endif | |
154 | return this_cpu; | |
155 | } | |
156 | ||
157 | /* | |
158 | * With HIGHRES=y we do not migrate the timer when it is expiring | |
159 | * before the next event on the target cpu because we cannot reprogram | |
160 | * the target cpu hardware and we would cause it to fire late. | |
161 | * | |
162 | * Called with cpu_base->lock of target cpu held. | |
163 | */ | |
164 | static int | |
165 | hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base) | |
166 | { | |
167 | #ifdef CONFIG_HIGH_RES_TIMERS | |
168 | ktime_t expires; | |
169 | ||
170 | if (!new_base->cpu_base->hres_active) | |
171 | return 0; | |
172 | ||
173 | expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset); | |
174 | return expires.tv64 <= new_base->cpu_base->expires_next.tv64; | |
175 | #else | |
176 | return 0; | |
177 | #endif | |
178 | } | |
179 | ||
c0a31329 TG |
180 | /* |
181 | * Switch the timer base to the current CPU when possible. | |
182 | */ | |
3c8aa39d | 183 | static inline struct hrtimer_clock_base * |
597d0275 AB |
184 | switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base, |
185 | int pinned) | |
c0a31329 | 186 | { |
3c8aa39d TG |
187 | struct hrtimer_clock_base *new_base; |
188 | struct hrtimer_cpu_base *new_cpu_base; | |
6ff7041d TG |
189 | int this_cpu = smp_processor_id(); |
190 | int cpu = hrtimer_get_target(this_cpu, pinned); | |
c0a31329 | 191 | |
eea08f32 AB |
192 | again: |
193 | new_cpu_base = &per_cpu(hrtimer_bases, cpu); | |
3c8aa39d | 194 | new_base = &new_cpu_base->clock_base[base->index]; |
c0a31329 TG |
195 | |
196 | if (base != new_base) { | |
197 | /* | |
6ff7041d | 198 | * We are trying to move timer to new_base. |
c0a31329 TG |
199 | * However we can't change timer's base while it is running, |
200 | * so we keep it on the same CPU. No hassle vs. reprogramming | |
201 | * the event source in the high resolution case. The softirq | |
202 | * code will take care of this when the timer function has | |
203 | * completed. There is no conflict as we hold the lock until | |
204 | * the timer is enqueued. | |
205 | */ | |
54cdfdb4 | 206 | if (unlikely(hrtimer_callback_running(timer))) |
c0a31329 TG |
207 | return base; |
208 | ||
209 | /* See the comment in lock_timer_base() */ | |
210 | timer->base = NULL; | |
3c8aa39d TG |
211 | spin_unlock(&base->cpu_base->lock); |
212 | spin_lock(&new_base->cpu_base->lock); | |
eea08f32 | 213 | |
6ff7041d TG |
214 | if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) { |
215 | cpu = this_cpu; | |
216 | spin_unlock(&new_base->cpu_base->lock); | |
217 | spin_lock(&base->cpu_base->lock); | |
218 | timer->base = base; | |
219 | goto again; | |
eea08f32 | 220 | } |
c0a31329 TG |
221 | timer->base = new_base; |
222 | } | |
223 | return new_base; | |
224 | } | |
225 | ||
226 | #else /* CONFIG_SMP */ | |
227 | ||
3c8aa39d | 228 | static inline struct hrtimer_clock_base * |
c0a31329 TG |
229 | lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) |
230 | { | |
3c8aa39d | 231 | struct hrtimer_clock_base *base = timer->base; |
c0a31329 | 232 | |
3c8aa39d | 233 | spin_lock_irqsave(&base->cpu_base->lock, *flags); |
c0a31329 TG |
234 | |
235 | return base; | |
236 | } | |
237 | ||
eea08f32 | 238 | # define switch_hrtimer_base(t, b, p) (b) |
c0a31329 TG |
239 | |
240 | #endif /* !CONFIG_SMP */ | |
241 | ||
242 | /* | |
243 | * Functions for the union type storage format of ktime_t which are | |
244 | * too large for inlining: | |
245 | */ | |
246 | #if BITS_PER_LONG < 64 | |
247 | # ifndef CONFIG_KTIME_SCALAR | |
248 | /** | |
249 | * ktime_add_ns - Add a scalar nanoseconds value to a ktime_t variable | |
c0a31329 TG |
250 | * @kt: addend |
251 | * @nsec: the scalar nsec value to add | |
252 | * | |
253 | * Returns the sum of kt and nsec in ktime_t format | |
254 | */ | |
255 | ktime_t ktime_add_ns(const ktime_t kt, u64 nsec) | |
256 | { | |
257 | ktime_t tmp; | |
258 | ||
259 | if (likely(nsec < NSEC_PER_SEC)) { | |
260 | tmp.tv64 = nsec; | |
261 | } else { | |
262 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
263 | ||
264 | tmp = ktime_set((long)nsec, rem); | |
265 | } | |
266 | ||
267 | return ktime_add(kt, tmp); | |
268 | } | |
b8b8fd2d DH |
269 | |
270 | EXPORT_SYMBOL_GPL(ktime_add_ns); | |
a272378d ACM |
271 | |
272 | /** | |
273 | * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable | |
274 | * @kt: minuend | |
275 | * @nsec: the scalar nsec value to subtract | |
276 | * | |
277 | * Returns the subtraction of @nsec from @kt in ktime_t format | |
278 | */ | |
279 | ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec) | |
280 | { | |
281 | ktime_t tmp; | |
282 | ||
283 | if (likely(nsec < NSEC_PER_SEC)) { | |
284 | tmp.tv64 = nsec; | |
285 | } else { | |
286 | unsigned long rem = do_div(nsec, NSEC_PER_SEC); | |
287 | ||
288 | tmp = ktime_set((long)nsec, rem); | |
289 | } | |
290 | ||
291 | return ktime_sub(kt, tmp); | |
292 | } | |
293 | ||
294 | EXPORT_SYMBOL_GPL(ktime_sub_ns); | |
c0a31329 TG |
295 | # endif /* !CONFIG_KTIME_SCALAR */ |
296 | ||
297 | /* | |
298 | * Divide a ktime value by a nanosecond value | |
299 | */ | |
4d672e7a | 300 | u64 ktime_divns(const ktime_t kt, s64 div) |
c0a31329 | 301 | { |
900cfa46 | 302 | u64 dclc; |
c0a31329 TG |
303 | int sft = 0; |
304 | ||
900cfa46 | 305 | dclc = ktime_to_ns(kt); |
c0a31329 TG |
306 | /* Make sure the divisor is less than 2^32: */ |
307 | while (div >> 32) { | |
308 | sft++; | |
309 | div >>= 1; | |
310 | } | |
311 | dclc >>= sft; | |
312 | do_div(dclc, (unsigned long) div); | |
313 | ||
4d672e7a | 314 | return dclc; |
c0a31329 | 315 | } |
c0a31329 TG |
316 | #endif /* BITS_PER_LONG >= 64 */ |
317 | ||
5a7780e7 TG |
318 | /* |
319 | * Add two ktime values and do a safety check for overflow: | |
320 | */ | |
321 | ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs) | |
322 | { | |
323 | ktime_t res = ktime_add(lhs, rhs); | |
324 | ||
325 | /* | |
326 | * We use KTIME_SEC_MAX here, the maximum timeout which we can | |
327 | * return to user space in a timespec: | |
328 | */ | |
329 | if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64) | |
330 | res = ktime_set(KTIME_SEC_MAX, 0); | |
331 | ||
332 | return res; | |
333 | } | |
334 | ||
8daa21e6 AB |
335 | EXPORT_SYMBOL_GPL(ktime_add_safe); |
336 | ||
237fc6e7 TG |
337 | #ifdef CONFIG_DEBUG_OBJECTS_TIMERS |
338 | ||
339 | static struct debug_obj_descr hrtimer_debug_descr; | |
340 | ||
341 | /* | |
342 | * fixup_init is called when: | |
343 | * - an active object is initialized | |
344 | */ | |
345 | static int hrtimer_fixup_init(void *addr, enum debug_obj_state state) | |
346 | { | |
347 | struct hrtimer *timer = addr; | |
348 | ||
349 | switch (state) { | |
350 | case ODEBUG_STATE_ACTIVE: | |
351 | hrtimer_cancel(timer); | |
352 | debug_object_init(timer, &hrtimer_debug_descr); | |
353 | return 1; | |
354 | default: | |
355 | return 0; | |
356 | } | |
357 | } | |
358 | ||
359 | /* | |
360 | * fixup_activate is called when: | |
361 | * - an active object is activated | |
362 | * - an unknown object is activated (might be a statically initialized object) | |
363 | */ | |
364 | static int hrtimer_fixup_activate(void *addr, enum debug_obj_state state) | |
365 | { | |
366 | switch (state) { | |
367 | ||
368 | case ODEBUG_STATE_NOTAVAILABLE: | |
369 | WARN_ON_ONCE(1); | |
370 | return 0; | |
371 | ||
372 | case ODEBUG_STATE_ACTIVE: | |
373 | WARN_ON(1); | |
374 | ||
375 | default: | |
376 | return 0; | |
377 | } | |
378 | } | |
379 | ||
380 | /* | |
381 | * fixup_free is called when: | |
382 | * - an active object is freed | |
383 | */ | |
384 | static int hrtimer_fixup_free(void *addr, enum debug_obj_state state) | |
385 | { | |
386 | struct hrtimer *timer = addr; | |
387 | ||
388 | switch (state) { | |
389 | case ODEBUG_STATE_ACTIVE: | |
390 | hrtimer_cancel(timer); | |
391 | debug_object_free(timer, &hrtimer_debug_descr); | |
392 | return 1; | |
393 | default: | |
394 | return 0; | |
395 | } | |
396 | } | |
397 | ||
398 | static struct debug_obj_descr hrtimer_debug_descr = { | |
399 | .name = "hrtimer", | |
400 | .fixup_init = hrtimer_fixup_init, | |
401 | .fixup_activate = hrtimer_fixup_activate, | |
402 | .fixup_free = hrtimer_fixup_free, | |
403 | }; | |
404 | ||
405 | static inline void debug_hrtimer_init(struct hrtimer *timer) | |
406 | { | |
407 | debug_object_init(timer, &hrtimer_debug_descr); | |
408 | } | |
409 | ||
410 | static inline void debug_hrtimer_activate(struct hrtimer *timer) | |
411 | { | |
412 | debug_object_activate(timer, &hrtimer_debug_descr); | |
413 | } | |
414 | ||
415 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) | |
416 | { | |
417 | debug_object_deactivate(timer, &hrtimer_debug_descr); | |
418 | } | |
419 | ||
420 | static inline void debug_hrtimer_free(struct hrtimer *timer) | |
421 | { | |
422 | debug_object_free(timer, &hrtimer_debug_descr); | |
423 | } | |
424 | ||
425 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
426 | enum hrtimer_mode mode); | |
427 | ||
428 | void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id, | |
429 | enum hrtimer_mode mode) | |
430 | { | |
431 | debug_object_init_on_stack(timer, &hrtimer_debug_descr); | |
432 | __hrtimer_init(timer, clock_id, mode); | |
433 | } | |
2bc481cf | 434 | EXPORT_SYMBOL_GPL(hrtimer_init_on_stack); |
237fc6e7 TG |
435 | |
436 | void destroy_hrtimer_on_stack(struct hrtimer *timer) | |
437 | { | |
438 | debug_object_free(timer, &hrtimer_debug_descr); | |
439 | } | |
440 | ||
441 | #else | |
442 | static inline void debug_hrtimer_init(struct hrtimer *timer) { } | |
443 | static inline void debug_hrtimer_activate(struct hrtimer *timer) { } | |
444 | static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { } | |
445 | #endif | |
446 | ||
c6a2a177 XG |
447 | static inline void |
448 | debug_init(struct hrtimer *timer, clockid_t clockid, | |
449 | enum hrtimer_mode mode) | |
450 | { | |
451 | debug_hrtimer_init(timer); | |
452 | trace_hrtimer_init(timer, clockid, mode); | |
453 | } | |
454 | ||
455 | static inline void debug_activate(struct hrtimer *timer) | |
456 | { | |
457 | debug_hrtimer_activate(timer); | |
458 | trace_hrtimer_start(timer); | |
459 | } | |
460 | ||
461 | static inline void debug_deactivate(struct hrtimer *timer) | |
462 | { | |
463 | debug_hrtimer_deactivate(timer); | |
464 | trace_hrtimer_cancel(timer); | |
465 | } | |
466 | ||
54cdfdb4 TG |
467 | /* High resolution timer related functions */ |
468 | #ifdef CONFIG_HIGH_RES_TIMERS | |
469 | ||
470 | /* | |
471 | * High resolution timer enabled ? | |
472 | */ | |
473 | static int hrtimer_hres_enabled __read_mostly = 1; | |
474 | ||
475 | /* | |
476 | * Enable / Disable high resolution mode | |
477 | */ | |
478 | static int __init setup_hrtimer_hres(char *str) | |
479 | { | |
480 | if (!strcmp(str, "off")) | |
481 | hrtimer_hres_enabled = 0; | |
482 | else if (!strcmp(str, "on")) | |
483 | hrtimer_hres_enabled = 1; | |
484 | else | |
485 | return 0; | |
486 | return 1; | |
487 | } | |
488 | ||
489 | __setup("highres=", setup_hrtimer_hres); | |
490 | ||
491 | /* | |
492 | * hrtimer_high_res_enabled - query, if the highres mode is enabled | |
493 | */ | |
494 | static inline int hrtimer_is_hres_enabled(void) | |
495 | { | |
496 | return hrtimer_hres_enabled; | |
497 | } | |
498 | ||
499 | /* | |
500 | * Is the high resolution mode active ? | |
501 | */ | |
502 | static inline int hrtimer_hres_active(void) | |
503 | { | |
504 | return __get_cpu_var(hrtimer_bases).hres_active; | |
505 | } | |
506 | ||
507 | /* | |
508 | * Reprogram the event source with checking both queues for the | |
509 | * next event | |
510 | * Called with interrupts disabled and base->lock held | |
511 | */ | |
512 | static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base) | |
513 | { | |
514 | int i; | |
515 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
516 | ktime_t expires; | |
517 | ||
518 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
519 | ||
520 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
521 | struct hrtimer *timer; | |
522 | ||
523 | if (!base->first) | |
524 | continue; | |
525 | timer = rb_entry(base->first, struct hrtimer, node); | |
cc584b21 | 526 | expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
b0a9b511 TG |
527 | /* |
528 | * clock_was_set() has changed base->offset so the | |
529 | * result might be negative. Fix it up to prevent a | |
530 | * false positive in clockevents_program_event() | |
531 | */ | |
532 | if (expires.tv64 < 0) | |
533 | expires.tv64 = 0; | |
54cdfdb4 TG |
534 | if (expires.tv64 < cpu_base->expires_next.tv64) |
535 | cpu_base->expires_next = expires; | |
536 | } | |
537 | ||
538 | if (cpu_base->expires_next.tv64 != KTIME_MAX) | |
539 | tick_program_event(cpu_base->expires_next, 1); | |
540 | } | |
541 | ||
542 | /* | |
543 | * Shared reprogramming for clock_realtime and clock_monotonic | |
544 | * | |
545 | * When a timer is enqueued and expires earlier than the already enqueued | |
546 | * timers, we have to check, whether it expires earlier than the timer for | |
547 | * which the clock event device was armed. | |
548 | * | |
549 | * Called with interrupts disabled and base->cpu_base.lock held | |
550 | */ | |
551 | static int hrtimer_reprogram(struct hrtimer *timer, | |
552 | struct hrtimer_clock_base *base) | |
553 | { | |
554 | ktime_t *expires_next = &__get_cpu_var(hrtimer_bases).expires_next; | |
cc584b21 | 555 | ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset); |
54cdfdb4 TG |
556 | int res; |
557 | ||
cc584b21 | 558 | WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0); |
63070a79 | 559 | |
54cdfdb4 TG |
560 | /* |
561 | * When the callback is running, we do not reprogram the clock event | |
562 | * device. The timer callback is either running on a different CPU or | |
3a4fa0a2 | 563 | * the callback is executed in the hrtimer_interrupt context. The |
54cdfdb4 TG |
564 | * reprogramming is handled either by the softirq, which called the |
565 | * callback or at the end of the hrtimer_interrupt. | |
566 | */ | |
567 | if (hrtimer_callback_running(timer)) | |
568 | return 0; | |
569 | ||
63070a79 TG |
570 | /* |
571 | * CLOCK_REALTIME timer might be requested with an absolute | |
572 | * expiry time which is less than base->offset. Nothing wrong | |
573 | * about that, just avoid to call into the tick code, which | |
574 | * has now objections against negative expiry values. | |
575 | */ | |
576 | if (expires.tv64 < 0) | |
577 | return -ETIME; | |
578 | ||
54cdfdb4 TG |
579 | if (expires.tv64 >= expires_next->tv64) |
580 | return 0; | |
581 | ||
582 | /* | |
583 | * Clockevents returns -ETIME, when the event was in the past. | |
584 | */ | |
585 | res = tick_program_event(expires, 0); | |
586 | if (!IS_ERR_VALUE(res)) | |
587 | *expires_next = expires; | |
588 | return res; | |
589 | } | |
590 | ||
591 | ||
592 | /* | |
593 | * Retrigger next event is called after clock was set | |
594 | * | |
595 | * Called with interrupts disabled via on_each_cpu() | |
596 | */ | |
597 | static void retrigger_next_event(void *arg) | |
598 | { | |
599 | struct hrtimer_cpu_base *base; | |
600 | struct timespec realtime_offset; | |
601 | unsigned long seq; | |
602 | ||
603 | if (!hrtimer_hres_active()) | |
604 | return; | |
605 | ||
606 | do { | |
607 | seq = read_seqbegin(&xtime_lock); | |
608 | set_normalized_timespec(&realtime_offset, | |
609 | -wall_to_monotonic.tv_sec, | |
610 | -wall_to_monotonic.tv_nsec); | |
611 | } while (read_seqretry(&xtime_lock, seq)); | |
612 | ||
613 | base = &__get_cpu_var(hrtimer_bases); | |
614 | ||
615 | /* Adjust CLOCK_REALTIME offset */ | |
616 | spin_lock(&base->lock); | |
617 | base->clock_base[CLOCK_REALTIME].offset = | |
618 | timespec_to_ktime(realtime_offset); | |
619 | ||
620 | hrtimer_force_reprogram(base); | |
621 | spin_unlock(&base->lock); | |
622 | } | |
623 | ||
624 | /* | |
625 | * Clock realtime was set | |
626 | * | |
627 | * Change the offset of the realtime clock vs. the monotonic | |
628 | * clock. | |
629 | * | |
630 | * We might have to reprogram the high resolution timer interrupt. On | |
631 | * SMP we call the architecture specific code to retrigger _all_ high | |
632 | * resolution timer interrupts. On UP we just disable interrupts and | |
633 | * call the high resolution interrupt code. | |
634 | */ | |
635 | void clock_was_set(void) | |
636 | { | |
637 | /* Retrigger the CPU local events everywhere */ | |
15c8b6c1 | 638 | on_each_cpu(retrigger_next_event, NULL, 1); |
54cdfdb4 TG |
639 | } |
640 | ||
995f054f IM |
641 | /* |
642 | * During resume we might have to reprogram the high resolution timer | |
643 | * interrupt (on the local CPU): | |
644 | */ | |
645 | void hres_timers_resume(void) | |
646 | { | |
1d4a7f1c PZ |
647 | WARN_ONCE(!irqs_disabled(), |
648 | KERN_INFO "hres_timers_resume() called with IRQs enabled!"); | |
649 | ||
995f054f IM |
650 | retrigger_next_event(NULL); |
651 | } | |
652 | ||
54cdfdb4 TG |
653 | /* |
654 | * Initialize the high resolution related parts of cpu_base | |
655 | */ | |
656 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) | |
657 | { | |
658 | base->expires_next.tv64 = KTIME_MAX; | |
659 | base->hres_active = 0; | |
54cdfdb4 TG |
660 | } |
661 | ||
662 | /* | |
663 | * Initialize the high resolution related parts of a hrtimer | |
664 | */ | |
665 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) | |
666 | { | |
54cdfdb4 TG |
667 | } |
668 | ||
ca109491 | 669 | |
54cdfdb4 TG |
670 | /* |
671 | * When High resolution timers are active, try to reprogram. Note, that in case | |
672 | * the state has HRTIMER_STATE_CALLBACK set, no reprogramming and no expiry | |
673 | * check happens. The timer gets enqueued into the rbtree. The reprogramming | |
674 | * and expiry check is done in the hrtimer_interrupt or in the softirq. | |
675 | */ | |
676 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
7f1e2ca9 PZ |
677 | struct hrtimer_clock_base *base, |
678 | int wakeup) | |
54cdfdb4 TG |
679 | { |
680 | if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) { | |
7f1e2ca9 PZ |
681 | if (wakeup) { |
682 | spin_unlock(&base->cpu_base->lock); | |
683 | raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
684 | spin_lock(&base->cpu_base->lock); | |
685 | } else | |
686 | __raise_softirq_irqoff(HRTIMER_SOFTIRQ); | |
687 | ||
ca109491 | 688 | return 1; |
54cdfdb4 | 689 | } |
7f1e2ca9 | 690 | |
54cdfdb4 TG |
691 | return 0; |
692 | } | |
693 | ||
694 | /* | |
695 | * Switch to high resolution mode | |
696 | */ | |
f8953856 | 697 | static int hrtimer_switch_to_hres(void) |
54cdfdb4 | 698 | { |
820de5c3 IM |
699 | int cpu = smp_processor_id(); |
700 | struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu); | |
54cdfdb4 TG |
701 | unsigned long flags; |
702 | ||
703 | if (base->hres_active) | |
f8953856 | 704 | return 1; |
54cdfdb4 TG |
705 | |
706 | local_irq_save(flags); | |
707 | ||
708 | if (tick_init_highres()) { | |
709 | local_irq_restore(flags); | |
820de5c3 IM |
710 | printk(KERN_WARNING "Could not switch to high resolution " |
711 | "mode on CPU %d\n", cpu); | |
f8953856 | 712 | return 0; |
54cdfdb4 TG |
713 | } |
714 | base->hres_active = 1; | |
715 | base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES; | |
716 | base->clock_base[CLOCK_MONOTONIC].resolution = KTIME_HIGH_RES; | |
717 | ||
718 | tick_setup_sched_timer(); | |
719 | ||
720 | /* "Retrigger" the interrupt to get things going */ | |
721 | retrigger_next_event(NULL); | |
722 | local_irq_restore(flags); | |
edfed66e | 723 | printk(KERN_DEBUG "Switched to high resolution mode on CPU %d\n", |
54cdfdb4 | 724 | smp_processor_id()); |
f8953856 | 725 | return 1; |
54cdfdb4 TG |
726 | } |
727 | ||
728 | #else | |
729 | ||
730 | static inline int hrtimer_hres_active(void) { return 0; } | |
731 | static inline int hrtimer_is_hres_enabled(void) { return 0; } | |
f8953856 | 732 | static inline int hrtimer_switch_to_hres(void) { return 0; } |
54cdfdb4 TG |
733 | static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { } |
734 | static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer, | |
7f1e2ca9 PZ |
735 | struct hrtimer_clock_base *base, |
736 | int wakeup) | |
54cdfdb4 TG |
737 | { |
738 | return 0; | |
739 | } | |
54cdfdb4 TG |
740 | static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { } |
741 | static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { } | |
742 | ||
743 | #endif /* CONFIG_HIGH_RES_TIMERS */ | |
744 | ||
82f67cd9 IM |
745 | #ifdef CONFIG_TIMER_STATS |
746 | void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr) | |
747 | { | |
748 | if (timer->start_site) | |
749 | return; | |
750 | ||
751 | timer->start_site = addr; | |
752 | memcpy(timer->start_comm, current->comm, TASK_COMM_LEN); | |
753 | timer->start_pid = current->pid; | |
754 | } | |
755 | #endif | |
756 | ||
c0a31329 | 757 | /* |
6506f2aa | 758 | * Counterpart to lock_hrtimer_base above: |
c0a31329 TG |
759 | */ |
760 | static inline | |
761 | void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags) | |
762 | { | |
3c8aa39d | 763 | spin_unlock_irqrestore(&timer->base->cpu_base->lock, *flags); |
c0a31329 TG |
764 | } |
765 | ||
766 | /** | |
767 | * hrtimer_forward - forward the timer expiry | |
c0a31329 | 768 | * @timer: hrtimer to forward |
44f21475 | 769 | * @now: forward past this time |
c0a31329 TG |
770 | * @interval: the interval to forward |
771 | * | |
772 | * Forward the timer expiry so it will expire in the future. | |
8dca6f33 | 773 | * Returns the number of overruns. |
c0a31329 | 774 | */ |
4d672e7a | 775 | u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval) |
c0a31329 | 776 | { |
4d672e7a | 777 | u64 orun = 1; |
44f21475 | 778 | ktime_t delta; |
c0a31329 | 779 | |
cc584b21 | 780 | delta = ktime_sub(now, hrtimer_get_expires(timer)); |
c0a31329 TG |
781 | |
782 | if (delta.tv64 < 0) | |
783 | return 0; | |
784 | ||
c9db4fa1 TG |
785 | if (interval.tv64 < timer->base->resolution.tv64) |
786 | interval.tv64 = timer->base->resolution.tv64; | |
787 | ||
c0a31329 | 788 | if (unlikely(delta.tv64 >= interval.tv64)) { |
df869b63 | 789 | s64 incr = ktime_to_ns(interval); |
c0a31329 TG |
790 | |
791 | orun = ktime_divns(delta, incr); | |
cc584b21 AV |
792 | hrtimer_add_expires_ns(timer, incr * orun); |
793 | if (hrtimer_get_expires_tv64(timer) > now.tv64) | |
c0a31329 TG |
794 | return orun; |
795 | /* | |
796 | * This (and the ktime_add() below) is the | |
797 | * correction for exact: | |
798 | */ | |
799 | orun++; | |
800 | } | |
cc584b21 | 801 | hrtimer_add_expires(timer, interval); |
c0a31329 TG |
802 | |
803 | return orun; | |
804 | } | |
6bdb6b62 | 805 | EXPORT_SYMBOL_GPL(hrtimer_forward); |
c0a31329 TG |
806 | |
807 | /* | |
808 | * enqueue_hrtimer - internal function to (re)start a timer | |
809 | * | |
810 | * The timer is inserted in expiry order. Insertion into the | |
811 | * red black tree is O(log(n)). Must hold the base lock. | |
a6037b61 PZ |
812 | * |
813 | * Returns 1 when the new timer is the leftmost timer in the tree. | |
c0a31329 | 814 | */ |
a6037b61 PZ |
815 | static int enqueue_hrtimer(struct hrtimer *timer, |
816 | struct hrtimer_clock_base *base) | |
c0a31329 TG |
817 | { |
818 | struct rb_node **link = &base->active.rb_node; | |
c0a31329 TG |
819 | struct rb_node *parent = NULL; |
820 | struct hrtimer *entry; | |
99bc2fcb | 821 | int leftmost = 1; |
c0a31329 | 822 | |
c6a2a177 | 823 | debug_activate(timer); |
237fc6e7 | 824 | |
c0a31329 TG |
825 | /* |
826 | * Find the right place in the rbtree: | |
827 | */ | |
828 | while (*link) { | |
829 | parent = *link; | |
830 | entry = rb_entry(parent, struct hrtimer, node); | |
831 | /* | |
832 | * We dont care about collisions. Nodes with | |
833 | * the same expiry time stay together. | |
834 | */ | |
cc584b21 AV |
835 | if (hrtimer_get_expires_tv64(timer) < |
836 | hrtimer_get_expires_tv64(entry)) { | |
c0a31329 | 837 | link = &(*link)->rb_left; |
99bc2fcb | 838 | } else { |
c0a31329 | 839 | link = &(*link)->rb_right; |
99bc2fcb IM |
840 | leftmost = 0; |
841 | } | |
c0a31329 TG |
842 | } |
843 | ||
844 | /* | |
288867ec TG |
845 | * Insert the timer to the rbtree and check whether it |
846 | * replaces the first pending timer | |
c0a31329 | 847 | */ |
a6037b61 | 848 | if (leftmost) |
54cdfdb4 | 849 | base->first = &timer->node; |
54cdfdb4 | 850 | |
c0a31329 TG |
851 | rb_link_node(&timer->node, parent, link); |
852 | rb_insert_color(&timer->node, &base->active); | |
303e967f TG |
853 | /* |
854 | * HRTIMER_STATE_ENQUEUED is or'ed to the current state to preserve the | |
855 | * state of a possibly running callback. | |
856 | */ | |
857 | timer->state |= HRTIMER_STATE_ENQUEUED; | |
a6037b61 PZ |
858 | |
859 | return leftmost; | |
288867ec | 860 | } |
c0a31329 TG |
861 | |
862 | /* | |
863 | * __remove_hrtimer - internal function to remove a timer | |
864 | * | |
865 | * Caller must hold the base lock. | |
54cdfdb4 TG |
866 | * |
867 | * High resolution timer mode reprograms the clock event device when the | |
868 | * timer is the one which expires next. The caller can disable this by setting | |
869 | * reprogram to zero. This is useful, when the context does a reprogramming | |
870 | * anyway (e.g. timer interrupt) | |
c0a31329 | 871 | */ |
3c8aa39d | 872 | static void __remove_hrtimer(struct hrtimer *timer, |
303e967f | 873 | struct hrtimer_clock_base *base, |
54cdfdb4 | 874 | unsigned long newstate, int reprogram) |
c0a31329 | 875 | { |
ca109491 | 876 | if (timer->state & HRTIMER_STATE_ENQUEUED) { |
54cdfdb4 TG |
877 | /* |
878 | * Remove the timer from the rbtree and replace the | |
879 | * first entry pointer if necessary. | |
880 | */ | |
881 | if (base->first == &timer->node) { | |
882 | base->first = rb_next(&timer->node); | |
883 | /* Reprogram the clock event device. if enabled */ | |
884 | if (reprogram && hrtimer_hres_active()) | |
885 | hrtimer_force_reprogram(base->cpu_base); | |
886 | } | |
887 | rb_erase(&timer->node, &base->active); | |
888 | } | |
303e967f | 889 | timer->state = newstate; |
c0a31329 TG |
890 | } |
891 | ||
892 | /* | |
893 | * remove hrtimer, called with base lock held | |
894 | */ | |
895 | static inline int | |
3c8aa39d | 896 | remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) |
c0a31329 | 897 | { |
303e967f | 898 | if (hrtimer_is_queued(timer)) { |
54cdfdb4 TG |
899 | int reprogram; |
900 | ||
901 | /* | |
902 | * Remove the timer and force reprogramming when high | |
903 | * resolution mode is active and the timer is on the current | |
904 | * CPU. If we remove a timer on another CPU, reprogramming is | |
905 | * skipped. The interrupt event on this CPU is fired and | |
906 | * reprogramming happens in the interrupt handler. This is a | |
907 | * rare case and less expensive than a smp call. | |
908 | */ | |
c6a2a177 | 909 | debug_deactivate(timer); |
82f67cd9 | 910 | timer_stats_hrtimer_clear_start_info(timer); |
54cdfdb4 TG |
911 | reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); |
912 | __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, | |
913 | reprogram); | |
c0a31329 TG |
914 | return 1; |
915 | } | |
916 | return 0; | |
917 | } | |
918 | ||
7f1e2ca9 PZ |
919 | int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, |
920 | unsigned long delta_ns, const enum hrtimer_mode mode, | |
921 | int wakeup) | |
c0a31329 | 922 | { |
3c8aa39d | 923 | struct hrtimer_clock_base *base, *new_base; |
c0a31329 | 924 | unsigned long flags; |
a6037b61 | 925 | int ret, leftmost; |
c0a31329 TG |
926 | |
927 | base = lock_hrtimer_base(timer, &flags); | |
928 | ||
929 | /* Remove an active timer from the queue: */ | |
930 | ret = remove_hrtimer(timer, base); | |
931 | ||
932 | /* Switch the timer base, if necessary: */ | |
597d0275 | 933 | new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED); |
c0a31329 | 934 | |
597d0275 | 935 | if (mode & HRTIMER_MODE_REL) { |
5a7780e7 | 936 | tim = ktime_add_safe(tim, new_base->get_time()); |
06027bdd IM |
937 | /* |
938 | * CONFIG_TIME_LOW_RES is a temporary way for architectures | |
939 | * to signal that they simply return xtime in | |
940 | * do_gettimeoffset(). In this case we want to round up by | |
941 | * resolution when starting a relative timer, to avoid short | |
942 | * timeouts. This will go away with the GTOD framework. | |
943 | */ | |
944 | #ifdef CONFIG_TIME_LOW_RES | |
5a7780e7 | 945 | tim = ktime_add_safe(tim, base->resolution); |
06027bdd IM |
946 | #endif |
947 | } | |
237fc6e7 | 948 | |
da8f2e17 | 949 | hrtimer_set_expires_range_ns(timer, tim, delta_ns); |
c0a31329 | 950 | |
82f67cd9 IM |
951 | timer_stats_hrtimer_set_start_info(timer); |
952 | ||
a6037b61 PZ |
953 | leftmost = enqueue_hrtimer(timer, new_base); |
954 | ||
935c631d IM |
955 | /* |
956 | * Only allow reprogramming if the new base is on this CPU. | |
957 | * (it might still be on another CPU if the timer was pending) | |
a6037b61 PZ |
958 | * |
959 | * XXX send_remote_softirq() ? | |
935c631d | 960 | */ |
a6037b61 | 961 | if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)) |
7f1e2ca9 | 962 | hrtimer_enqueue_reprogram(timer, new_base, wakeup); |
c0a31329 TG |
963 | |
964 | unlock_hrtimer_base(timer, &flags); | |
965 | ||
966 | return ret; | |
967 | } | |
7f1e2ca9 PZ |
968 | |
969 | /** | |
970 | * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU | |
971 | * @timer: the timer to be added | |
972 | * @tim: expiry time | |
973 | * @delta_ns: "slack" range for the timer | |
974 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
975 | * | |
976 | * Returns: | |
977 | * 0 on success | |
978 | * 1 when the timer was active | |
979 | */ | |
980 | int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, | |
981 | unsigned long delta_ns, const enum hrtimer_mode mode) | |
982 | { | |
983 | return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1); | |
984 | } | |
da8f2e17 AV |
985 | EXPORT_SYMBOL_GPL(hrtimer_start_range_ns); |
986 | ||
987 | /** | |
e1dd7bc5 | 988 | * hrtimer_start - (re)start an hrtimer on the current CPU |
da8f2e17 AV |
989 | * @timer: the timer to be added |
990 | * @tim: expiry time | |
991 | * @mode: expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL) | |
992 | * | |
993 | * Returns: | |
994 | * 0 on success | |
995 | * 1 when the timer was active | |
996 | */ | |
997 | int | |
998 | hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode) | |
999 | { | |
7f1e2ca9 | 1000 | return __hrtimer_start_range_ns(timer, tim, 0, mode, 1); |
da8f2e17 | 1001 | } |
8d16b764 | 1002 | EXPORT_SYMBOL_GPL(hrtimer_start); |
c0a31329 | 1003 | |
da8f2e17 | 1004 | |
c0a31329 TG |
1005 | /** |
1006 | * hrtimer_try_to_cancel - try to deactivate a timer | |
c0a31329 TG |
1007 | * @timer: hrtimer to stop |
1008 | * | |
1009 | * Returns: | |
1010 | * 0 when the timer was not active | |
1011 | * 1 when the timer was active | |
1012 | * -1 when the timer is currently excuting the callback function and | |
fa9799e3 | 1013 | * cannot be stopped |
c0a31329 TG |
1014 | */ |
1015 | int hrtimer_try_to_cancel(struct hrtimer *timer) | |
1016 | { | |
3c8aa39d | 1017 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1018 | unsigned long flags; |
1019 | int ret = -1; | |
1020 | ||
1021 | base = lock_hrtimer_base(timer, &flags); | |
1022 | ||
303e967f | 1023 | if (!hrtimer_callback_running(timer)) |
c0a31329 TG |
1024 | ret = remove_hrtimer(timer, base); |
1025 | ||
1026 | unlock_hrtimer_base(timer, &flags); | |
1027 | ||
1028 | return ret; | |
1029 | ||
1030 | } | |
8d16b764 | 1031 | EXPORT_SYMBOL_GPL(hrtimer_try_to_cancel); |
c0a31329 TG |
1032 | |
1033 | /** | |
1034 | * hrtimer_cancel - cancel a timer and wait for the handler to finish. | |
c0a31329 TG |
1035 | * @timer: the timer to be cancelled |
1036 | * | |
1037 | * Returns: | |
1038 | * 0 when the timer was not active | |
1039 | * 1 when the timer was active | |
1040 | */ | |
1041 | int hrtimer_cancel(struct hrtimer *timer) | |
1042 | { | |
1043 | for (;;) { | |
1044 | int ret = hrtimer_try_to_cancel(timer); | |
1045 | ||
1046 | if (ret >= 0) | |
1047 | return ret; | |
5ef37b19 | 1048 | cpu_relax(); |
c0a31329 TG |
1049 | } |
1050 | } | |
8d16b764 | 1051 | EXPORT_SYMBOL_GPL(hrtimer_cancel); |
c0a31329 TG |
1052 | |
1053 | /** | |
1054 | * hrtimer_get_remaining - get remaining time for the timer | |
c0a31329 TG |
1055 | * @timer: the timer to read |
1056 | */ | |
1057 | ktime_t hrtimer_get_remaining(const struct hrtimer *timer) | |
1058 | { | |
3c8aa39d | 1059 | struct hrtimer_clock_base *base; |
c0a31329 TG |
1060 | unsigned long flags; |
1061 | ktime_t rem; | |
1062 | ||
1063 | base = lock_hrtimer_base(timer, &flags); | |
cc584b21 | 1064 | rem = hrtimer_expires_remaining(timer); |
c0a31329 TG |
1065 | unlock_hrtimer_base(timer, &flags); |
1066 | ||
1067 | return rem; | |
1068 | } | |
8d16b764 | 1069 | EXPORT_SYMBOL_GPL(hrtimer_get_remaining); |
c0a31329 | 1070 | |
ee9c5785 | 1071 | #ifdef CONFIG_NO_HZ |
69239749 TL |
1072 | /** |
1073 | * hrtimer_get_next_event - get the time until next expiry event | |
1074 | * | |
1075 | * Returns the delta to the next expiry event or KTIME_MAX if no timer | |
1076 | * is pending. | |
1077 | */ | |
1078 | ktime_t hrtimer_get_next_event(void) | |
1079 | { | |
3c8aa39d TG |
1080 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1081 | struct hrtimer_clock_base *base = cpu_base->clock_base; | |
69239749 TL |
1082 | ktime_t delta, mindelta = { .tv64 = KTIME_MAX }; |
1083 | unsigned long flags; | |
1084 | int i; | |
1085 | ||
3c8aa39d TG |
1086 | spin_lock_irqsave(&cpu_base->lock, flags); |
1087 | ||
54cdfdb4 TG |
1088 | if (!hrtimer_hres_active()) { |
1089 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++, base++) { | |
1090 | struct hrtimer *timer; | |
69239749 | 1091 | |
54cdfdb4 TG |
1092 | if (!base->first) |
1093 | continue; | |
3c8aa39d | 1094 | |
54cdfdb4 | 1095 | timer = rb_entry(base->first, struct hrtimer, node); |
cc584b21 | 1096 | delta.tv64 = hrtimer_get_expires_tv64(timer); |
54cdfdb4 TG |
1097 | delta = ktime_sub(delta, base->get_time()); |
1098 | if (delta.tv64 < mindelta.tv64) | |
1099 | mindelta.tv64 = delta.tv64; | |
1100 | } | |
69239749 | 1101 | } |
3c8aa39d TG |
1102 | |
1103 | spin_unlock_irqrestore(&cpu_base->lock, flags); | |
1104 | ||
69239749 TL |
1105 | if (mindelta.tv64 < 0) |
1106 | mindelta.tv64 = 0; | |
1107 | return mindelta; | |
1108 | } | |
1109 | #endif | |
1110 | ||
237fc6e7 TG |
1111 | static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id, |
1112 | enum hrtimer_mode mode) | |
c0a31329 | 1113 | { |
3c8aa39d | 1114 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1115 | |
7978672c GA |
1116 | memset(timer, 0, sizeof(struct hrtimer)); |
1117 | ||
3c8aa39d | 1118 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
c0a31329 | 1119 | |
c9cb2e3d | 1120 | if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS) |
7978672c GA |
1121 | clock_id = CLOCK_MONOTONIC; |
1122 | ||
3c8aa39d | 1123 | timer->base = &cpu_base->clock_base[clock_id]; |
54cdfdb4 | 1124 | hrtimer_init_timer_hres(timer); |
82f67cd9 IM |
1125 | |
1126 | #ifdef CONFIG_TIMER_STATS | |
1127 | timer->start_site = NULL; | |
1128 | timer->start_pid = -1; | |
1129 | memset(timer->start_comm, 0, TASK_COMM_LEN); | |
1130 | #endif | |
c0a31329 | 1131 | } |
237fc6e7 TG |
1132 | |
1133 | /** | |
1134 | * hrtimer_init - initialize a timer to the given clock | |
1135 | * @timer: the timer to be initialized | |
1136 | * @clock_id: the clock to be used | |
1137 | * @mode: timer mode abs/rel | |
1138 | */ | |
1139 | void hrtimer_init(struct hrtimer *timer, clockid_t clock_id, | |
1140 | enum hrtimer_mode mode) | |
1141 | { | |
c6a2a177 | 1142 | debug_init(timer, clock_id, mode); |
237fc6e7 TG |
1143 | __hrtimer_init(timer, clock_id, mode); |
1144 | } | |
8d16b764 | 1145 | EXPORT_SYMBOL_GPL(hrtimer_init); |
c0a31329 TG |
1146 | |
1147 | /** | |
1148 | * hrtimer_get_res - get the timer resolution for a clock | |
c0a31329 TG |
1149 | * @which_clock: which clock to query |
1150 | * @tp: pointer to timespec variable to store the resolution | |
1151 | * | |
72fd4a35 RD |
1152 | * Store the resolution of the clock selected by @which_clock in the |
1153 | * variable pointed to by @tp. | |
c0a31329 TG |
1154 | */ |
1155 | int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp) | |
1156 | { | |
3c8aa39d | 1157 | struct hrtimer_cpu_base *cpu_base; |
c0a31329 | 1158 | |
3c8aa39d TG |
1159 | cpu_base = &__raw_get_cpu_var(hrtimer_bases); |
1160 | *tp = ktime_to_timespec(cpu_base->clock_base[which_clock].resolution); | |
c0a31329 TG |
1161 | |
1162 | return 0; | |
1163 | } | |
8d16b764 | 1164 | EXPORT_SYMBOL_GPL(hrtimer_get_res); |
c0a31329 | 1165 | |
c6a2a177 | 1166 | static void __run_hrtimer(struct hrtimer *timer, ktime_t *now) |
d3d74453 PZ |
1167 | { |
1168 | struct hrtimer_clock_base *base = timer->base; | |
1169 | struct hrtimer_cpu_base *cpu_base = base->cpu_base; | |
1170 | enum hrtimer_restart (*fn)(struct hrtimer *); | |
1171 | int restart; | |
1172 | ||
ca109491 PZ |
1173 | WARN_ON(!irqs_disabled()); |
1174 | ||
c6a2a177 | 1175 | debug_deactivate(timer); |
d3d74453 PZ |
1176 | __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0); |
1177 | timer_stats_account_hrtimer(timer); | |
d3d74453 | 1178 | fn = timer->function; |
ca109491 PZ |
1179 | |
1180 | /* | |
1181 | * Because we run timers from hardirq context, there is no chance | |
1182 | * they get migrated to another cpu, therefore its safe to unlock | |
1183 | * the timer base. | |
1184 | */ | |
1185 | spin_unlock(&cpu_base->lock); | |
c6a2a177 | 1186 | trace_hrtimer_expire_entry(timer, now); |
ca109491 | 1187 | restart = fn(timer); |
c6a2a177 | 1188 | trace_hrtimer_expire_exit(timer); |
ca109491 | 1189 | spin_lock(&cpu_base->lock); |
d3d74453 PZ |
1190 | |
1191 | /* | |
e3f1d883 TG |
1192 | * Note: We clear the CALLBACK bit after enqueue_hrtimer and |
1193 | * we do not reprogramm the event hardware. Happens either in | |
1194 | * hrtimer_start_range_ns() or in hrtimer_interrupt() | |
d3d74453 PZ |
1195 | */ |
1196 | if (restart != HRTIMER_NORESTART) { | |
1197 | BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); | |
a6037b61 | 1198 | enqueue_hrtimer(timer, base); |
d3d74453 PZ |
1199 | } |
1200 | timer->state &= ~HRTIMER_STATE_CALLBACK; | |
1201 | } | |
1202 | ||
54cdfdb4 TG |
1203 | #ifdef CONFIG_HIGH_RES_TIMERS |
1204 | ||
7f22391c FW |
1205 | static int force_clock_reprogram; |
1206 | ||
1207 | /* | |
1208 | * After 5 iteration's attempts, we consider that hrtimer_interrupt() | |
1209 | * is hanging, which could happen with something that slows the interrupt | |
1210 | * such as the tracing. Then we force the clock reprogramming for each future | |
1211 | * hrtimer interrupts to avoid infinite loops and use the min_delta_ns | |
1212 | * threshold that we will overwrite. | |
1213 | * The next tick event will be scheduled to 3 times we currently spend on | |
1214 | * hrtimer_interrupt(). This gives a good compromise, the cpus will spend | |
1215 | * 1/4 of their time to process the hrtimer interrupts. This is enough to | |
1216 | * let it running without serious starvation. | |
1217 | */ | |
1218 | ||
1219 | static inline void | |
1220 | hrtimer_interrupt_hanging(struct clock_event_device *dev, | |
1221 | ktime_t try_time) | |
1222 | { | |
1223 | force_clock_reprogram = 1; | |
1224 | dev->min_delta_ns = (unsigned long)try_time.tv64 * 3; | |
1225 | printk(KERN_WARNING "hrtimer: interrupt too slow, " | |
1226 | "forcing clock min delta to %lu ns\n", dev->min_delta_ns); | |
1227 | } | |
54cdfdb4 TG |
1228 | /* |
1229 | * High resolution timer interrupt | |
1230 | * Called with interrupts disabled | |
1231 | */ | |
1232 | void hrtimer_interrupt(struct clock_event_device *dev) | |
1233 | { | |
1234 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); | |
1235 | struct hrtimer_clock_base *base; | |
1236 | ktime_t expires_next, now; | |
7f22391c | 1237 | int nr_retries = 0; |
ca109491 | 1238 | int i; |
54cdfdb4 TG |
1239 | |
1240 | BUG_ON(!cpu_base->hres_active); | |
1241 | cpu_base->nr_events++; | |
1242 | dev->next_event.tv64 = KTIME_MAX; | |
1243 | ||
1244 | retry: | |
7f22391c FW |
1245 | /* 5 retries is enough to notice a hang */ |
1246 | if (!(++nr_retries % 5)) | |
1247 | hrtimer_interrupt_hanging(dev, ktime_sub(ktime_get(), now)); | |
1248 | ||
54cdfdb4 TG |
1249 | now = ktime_get(); |
1250 | ||
1251 | expires_next.tv64 = KTIME_MAX; | |
1252 | ||
6ff7041d TG |
1253 | spin_lock(&cpu_base->lock); |
1254 | /* | |
1255 | * We set expires_next to KTIME_MAX here with cpu_base->lock | |
1256 | * held to prevent that a timer is enqueued in our queue via | |
1257 | * the migration code. This does not affect enqueueing of | |
1258 | * timers which run their callback and need to be requeued on | |
1259 | * this CPU. | |
1260 | */ | |
1261 | cpu_base->expires_next.tv64 = KTIME_MAX; | |
1262 | ||
54cdfdb4 TG |
1263 | base = cpu_base->clock_base; |
1264 | ||
1265 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { | |
1266 | ktime_t basenow; | |
1267 | struct rb_node *node; | |
1268 | ||
54cdfdb4 TG |
1269 | basenow = ktime_add(now, base->offset); |
1270 | ||
1271 | while ((node = base->first)) { | |
1272 | struct hrtimer *timer; | |
1273 | ||
1274 | timer = rb_entry(node, struct hrtimer, node); | |
1275 | ||
654c8e0b AV |
1276 | /* |
1277 | * The immediate goal for using the softexpires is | |
1278 | * minimizing wakeups, not running timers at the | |
1279 | * earliest interrupt after their soft expiration. | |
1280 | * This allows us to avoid using a Priority Search | |
1281 | * Tree, which can answer a stabbing querry for | |
1282 | * overlapping intervals and instead use the simple | |
1283 | * BST we already have. | |
1284 | * We don't add extra wakeups by delaying timers that | |
1285 | * are right-of a not yet expired timer, because that | |
1286 | * timer will have to trigger a wakeup anyway. | |
1287 | */ | |
1288 | ||
1289 | if (basenow.tv64 < hrtimer_get_softexpires_tv64(timer)) { | |
54cdfdb4 TG |
1290 | ktime_t expires; |
1291 | ||
cc584b21 | 1292 | expires = ktime_sub(hrtimer_get_expires(timer), |
54cdfdb4 TG |
1293 | base->offset); |
1294 | if (expires.tv64 < expires_next.tv64) | |
1295 | expires_next = expires; | |
1296 | break; | |
1297 | } | |
1298 | ||
c6a2a177 | 1299 | __run_hrtimer(timer, &basenow); |
54cdfdb4 | 1300 | } |
54cdfdb4 TG |
1301 | base++; |
1302 | } | |
1303 | ||
6ff7041d TG |
1304 | /* |
1305 | * Store the new expiry value so the migration code can verify | |
1306 | * against it. | |
1307 | */ | |
54cdfdb4 | 1308 | cpu_base->expires_next = expires_next; |
6ff7041d | 1309 | spin_unlock(&cpu_base->lock); |
54cdfdb4 TG |
1310 | |
1311 | /* Reprogramming necessary ? */ | |
1312 | if (expires_next.tv64 != KTIME_MAX) { | |
7f22391c | 1313 | if (tick_program_event(expires_next, force_clock_reprogram)) |
54cdfdb4 TG |
1314 | goto retry; |
1315 | } | |
54cdfdb4 TG |
1316 | } |
1317 | ||
8bdec955 TG |
1318 | /* |
1319 | * local version of hrtimer_peek_ahead_timers() called with interrupts | |
1320 | * disabled. | |
1321 | */ | |
1322 | static void __hrtimer_peek_ahead_timers(void) | |
1323 | { | |
1324 | struct tick_device *td; | |
1325 | ||
1326 | if (!hrtimer_hres_active()) | |
1327 | return; | |
1328 | ||
1329 | td = &__get_cpu_var(tick_cpu_device); | |
1330 | if (td && td->evtdev) | |
1331 | hrtimer_interrupt(td->evtdev); | |
1332 | } | |
1333 | ||
2e94d1f7 AV |
1334 | /** |
1335 | * hrtimer_peek_ahead_timers -- run soft-expired timers now | |
1336 | * | |
1337 | * hrtimer_peek_ahead_timers will peek at the timer queue of | |
1338 | * the current cpu and check if there are any timers for which | |
1339 | * the soft expires time has passed. If any such timers exist, | |
1340 | * they are run immediately and then removed from the timer queue. | |
1341 | * | |
1342 | */ | |
1343 | void hrtimer_peek_ahead_timers(void) | |
1344 | { | |
643bdf68 | 1345 | unsigned long flags; |
dc4304f7 | 1346 | |
2e94d1f7 | 1347 | local_irq_save(flags); |
8bdec955 | 1348 | __hrtimer_peek_ahead_timers(); |
2e94d1f7 AV |
1349 | local_irq_restore(flags); |
1350 | } | |
1351 | ||
a6037b61 PZ |
1352 | static void run_hrtimer_softirq(struct softirq_action *h) |
1353 | { | |
1354 | hrtimer_peek_ahead_timers(); | |
1355 | } | |
1356 | ||
82c5b7b5 IM |
1357 | #else /* CONFIG_HIGH_RES_TIMERS */ |
1358 | ||
1359 | static inline void __hrtimer_peek_ahead_timers(void) { } | |
1360 | ||
1361 | #endif /* !CONFIG_HIGH_RES_TIMERS */ | |
82f67cd9 | 1362 | |
d3d74453 PZ |
1363 | /* |
1364 | * Called from timer softirq every jiffy, expire hrtimers: | |
1365 | * | |
1366 | * For HRT its the fall back code to run the softirq in the timer | |
1367 | * softirq context in case the hrtimer initialization failed or has | |
1368 | * not been done yet. | |
1369 | */ | |
1370 | void hrtimer_run_pending(void) | |
1371 | { | |
d3d74453 PZ |
1372 | if (hrtimer_hres_active()) |
1373 | return; | |
54cdfdb4 | 1374 | |
d3d74453 PZ |
1375 | /* |
1376 | * This _is_ ugly: We have to check in the softirq context, | |
1377 | * whether we can switch to highres and / or nohz mode. The | |
1378 | * clocksource switch happens in the timer interrupt with | |
1379 | * xtime_lock held. Notification from there only sets the | |
1380 | * check bit in the tick_oneshot code, otherwise we might | |
1381 | * deadlock vs. xtime_lock. | |
1382 | */ | |
1383 | if (tick_check_oneshot_change(!hrtimer_is_hres_enabled())) | |
1384 | hrtimer_switch_to_hres(); | |
54cdfdb4 TG |
1385 | } |
1386 | ||
c0a31329 | 1387 | /* |
d3d74453 | 1388 | * Called from hardirq context every jiffy |
c0a31329 | 1389 | */ |
833883d9 | 1390 | void hrtimer_run_queues(void) |
c0a31329 | 1391 | { |
288867ec | 1392 | struct rb_node *node; |
833883d9 DS |
1393 | struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases); |
1394 | struct hrtimer_clock_base *base; | |
1395 | int index, gettime = 1; | |
c0a31329 | 1396 | |
833883d9 | 1397 | if (hrtimer_hres_active()) |
3055adda DS |
1398 | return; |
1399 | ||
833883d9 DS |
1400 | for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { |
1401 | base = &cpu_base->clock_base[index]; | |
c0a31329 | 1402 | |
833883d9 | 1403 | if (!base->first) |
d3d74453 | 1404 | continue; |
833883d9 | 1405 | |
d7cfb60c | 1406 | if (gettime) { |
833883d9 DS |
1407 | hrtimer_get_softirq_time(cpu_base); |
1408 | gettime = 0; | |
b75f7a51 | 1409 | } |
d3d74453 | 1410 | |
833883d9 | 1411 | spin_lock(&cpu_base->lock); |
c0a31329 | 1412 | |
833883d9 DS |
1413 | while ((node = base->first)) { |
1414 | struct hrtimer *timer; | |
54cdfdb4 | 1415 | |
833883d9 | 1416 | timer = rb_entry(node, struct hrtimer, node); |
cc584b21 AV |
1417 | if (base->softirq_time.tv64 <= |
1418 | hrtimer_get_expires_tv64(timer)) | |
833883d9 DS |
1419 | break; |
1420 | ||
c6a2a177 | 1421 | __run_hrtimer(timer, &base->softirq_time); |
833883d9 DS |
1422 | } |
1423 | spin_unlock(&cpu_base->lock); | |
1424 | } | |
c0a31329 TG |
1425 | } |
1426 | ||
10c94ec1 TG |
1427 | /* |
1428 | * Sleep related functions: | |
1429 | */ | |
c9cb2e3d | 1430 | static enum hrtimer_restart hrtimer_wakeup(struct hrtimer *timer) |
00362e33 TG |
1431 | { |
1432 | struct hrtimer_sleeper *t = | |
1433 | container_of(timer, struct hrtimer_sleeper, timer); | |
1434 | struct task_struct *task = t->task; | |
1435 | ||
1436 | t->task = NULL; | |
1437 | if (task) | |
1438 | wake_up_process(task); | |
1439 | ||
1440 | return HRTIMER_NORESTART; | |
1441 | } | |
1442 | ||
36c8b586 | 1443 | void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task) |
00362e33 TG |
1444 | { |
1445 | sl->timer.function = hrtimer_wakeup; | |
1446 | sl->task = task; | |
1447 | } | |
2bc481cf | 1448 | EXPORT_SYMBOL_GPL(hrtimer_init_sleeper); |
00362e33 | 1449 | |
669d7868 | 1450 | static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode) |
432569bb | 1451 | { |
669d7868 | 1452 | hrtimer_init_sleeper(t, current); |
10c94ec1 | 1453 | |
432569bb RZ |
1454 | do { |
1455 | set_current_state(TASK_INTERRUPTIBLE); | |
cc584b21 | 1456 | hrtimer_start_expires(&t->timer, mode); |
37bb6cb4 PZ |
1457 | if (!hrtimer_active(&t->timer)) |
1458 | t->task = NULL; | |
432569bb | 1459 | |
54cdfdb4 TG |
1460 | if (likely(t->task)) |
1461 | schedule(); | |
432569bb | 1462 | |
669d7868 | 1463 | hrtimer_cancel(&t->timer); |
c9cb2e3d | 1464 | mode = HRTIMER_MODE_ABS; |
669d7868 TG |
1465 | |
1466 | } while (t->task && !signal_pending(current)); | |
432569bb | 1467 | |
3588a085 PZ |
1468 | __set_current_state(TASK_RUNNING); |
1469 | ||
669d7868 | 1470 | return t->task == NULL; |
10c94ec1 TG |
1471 | } |
1472 | ||
080344b9 ON |
1473 | static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp) |
1474 | { | |
1475 | struct timespec rmt; | |
1476 | ktime_t rem; | |
1477 | ||
cc584b21 | 1478 | rem = hrtimer_expires_remaining(timer); |
080344b9 ON |
1479 | if (rem.tv64 <= 0) |
1480 | return 0; | |
1481 | rmt = ktime_to_timespec(rem); | |
1482 | ||
1483 | if (copy_to_user(rmtp, &rmt, sizeof(*rmtp))) | |
1484 | return -EFAULT; | |
1485 | ||
1486 | return 1; | |
1487 | } | |
1488 | ||
1711ef38 | 1489 | long __sched hrtimer_nanosleep_restart(struct restart_block *restart) |
10c94ec1 | 1490 | { |
669d7868 | 1491 | struct hrtimer_sleeper t; |
080344b9 | 1492 | struct timespec __user *rmtp; |
237fc6e7 | 1493 | int ret = 0; |
10c94ec1 | 1494 | |
237fc6e7 TG |
1495 | hrtimer_init_on_stack(&t.timer, restart->nanosleep.index, |
1496 | HRTIMER_MODE_ABS); | |
cc584b21 | 1497 | hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires); |
10c94ec1 | 1498 | |
c9cb2e3d | 1499 | if (do_nanosleep(&t, HRTIMER_MODE_ABS)) |
237fc6e7 | 1500 | goto out; |
10c94ec1 | 1501 | |
029a07e0 | 1502 | rmtp = restart->nanosleep.rmtp; |
432569bb | 1503 | if (rmtp) { |
237fc6e7 | 1504 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1505 | if (ret <= 0) |
237fc6e7 | 1506 | goto out; |
432569bb | 1507 | } |
10c94ec1 | 1508 | |
10c94ec1 | 1509 | /* The other values in restart are already filled in */ |
237fc6e7 TG |
1510 | ret = -ERESTART_RESTARTBLOCK; |
1511 | out: | |
1512 | destroy_hrtimer_on_stack(&t.timer); | |
1513 | return ret; | |
10c94ec1 TG |
1514 | } |
1515 | ||
080344b9 | 1516 | long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp, |
10c94ec1 TG |
1517 | const enum hrtimer_mode mode, const clockid_t clockid) |
1518 | { | |
1519 | struct restart_block *restart; | |
669d7868 | 1520 | struct hrtimer_sleeper t; |
237fc6e7 | 1521 | int ret = 0; |
3bd01206 AV |
1522 | unsigned long slack; |
1523 | ||
1524 | slack = current->timer_slack_ns; | |
1525 | if (rt_task(current)) | |
1526 | slack = 0; | |
10c94ec1 | 1527 | |
237fc6e7 | 1528 | hrtimer_init_on_stack(&t.timer, clockid, mode); |
3bd01206 | 1529 | hrtimer_set_expires_range_ns(&t.timer, timespec_to_ktime(*rqtp), slack); |
432569bb | 1530 | if (do_nanosleep(&t, mode)) |
237fc6e7 | 1531 | goto out; |
10c94ec1 | 1532 | |
7978672c | 1533 | /* Absolute timers do not update the rmtp value and restart: */ |
237fc6e7 TG |
1534 | if (mode == HRTIMER_MODE_ABS) { |
1535 | ret = -ERESTARTNOHAND; | |
1536 | goto out; | |
1537 | } | |
10c94ec1 | 1538 | |
432569bb | 1539 | if (rmtp) { |
237fc6e7 | 1540 | ret = update_rmtp(&t.timer, rmtp); |
080344b9 | 1541 | if (ret <= 0) |
237fc6e7 | 1542 | goto out; |
432569bb | 1543 | } |
10c94ec1 TG |
1544 | |
1545 | restart = ¤t_thread_info()->restart_block; | |
1711ef38 | 1546 | restart->fn = hrtimer_nanosleep_restart; |
029a07e0 TG |
1547 | restart->nanosleep.index = t.timer.base->index; |
1548 | restart->nanosleep.rmtp = rmtp; | |
cc584b21 | 1549 | restart->nanosleep.expires = hrtimer_get_expires_tv64(&t.timer); |
10c94ec1 | 1550 | |
237fc6e7 TG |
1551 | ret = -ERESTART_RESTARTBLOCK; |
1552 | out: | |
1553 | destroy_hrtimer_on_stack(&t.timer); | |
1554 | return ret; | |
10c94ec1 TG |
1555 | } |
1556 | ||
58fd3aa2 HC |
1557 | SYSCALL_DEFINE2(nanosleep, struct timespec __user *, rqtp, |
1558 | struct timespec __user *, rmtp) | |
6ba1b912 | 1559 | { |
080344b9 | 1560 | struct timespec tu; |
6ba1b912 TG |
1561 | |
1562 | if (copy_from_user(&tu, rqtp, sizeof(tu))) | |
1563 | return -EFAULT; | |
1564 | ||
1565 | if (!timespec_valid(&tu)) | |
1566 | return -EINVAL; | |
1567 | ||
080344b9 | 1568 | return hrtimer_nanosleep(&tu, rmtp, HRTIMER_MODE_REL, CLOCK_MONOTONIC); |
6ba1b912 TG |
1569 | } |
1570 | ||
c0a31329 TG |
1571 | /* |
1572 | * Functions related to boot-time initialization: | |
1573 | */ | |
0ec160dd | 1574 | static void __cpuinit init_hrtimers_cpu(int cpu) |
c0a31329 | 1575 | { |
3c8aa39d | 1576 | struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu); |
c0a31329 TG |
1577 | int i; |
1578 | ||
3c8aa39d | 1579 | spin_lock_init(&cpu_base->lock); |
3c8aa39d TG |
1580 | |
1581 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) | |
1582 | cpu_base->clock_base[i].cpu_base = cpu_base; | |
1583 | ||
54cdfdb4 | 1584 | hrtimer_init_hres(cpu_base); |
c0a31329 TG |
1585 | } |
1586 | ||
1587 | #ifdef CONFIG_HOTPLUG_CPU | |
1588 | ||
ca109491 | 1589 | static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base, |
37810659 | 1590 | struct hrtimer_clock_base *new_base) |
c0a31329 TG |
1591 | { |
1592 | struct hrtimer *timer; | |
1593 | struct rb_node *node; | |
1594 | ||
1595 | while ((node = rb_first(&old_base->active))) { | |
1596 | timer = rb_entry(node, struct hrtimer, node); | |
54cdfdb4 | 1597 | BUG_ON(hrtimer_callback_running(timer)); |
c6a2a177 | 1598 | debug_deactivate(timer); |
b00c1a99 TG |
1599 | |
1600 | /* | |
1601 | * Mark it as STATE_MIGRATE not INACTIVE otherwise the | |
1602 | * timer could be seen as !active and just vanish away | |
1603 | * under us on another CPU | |
1604 | */ | |
1605 | __remove_hrtimer(timer, old_base, HRTIMER_STATE_MIGRATE, 0); | |
c0a31329 | 1606 | timer->base = new_base; |
54cdfdb4 | 1607 | /* |
e3f1d883 TG |
1608 | * Enqueue the timers on the new cpu. This does not |
1609 | * reprogram the event device in case the timer | |
1610 | * expires before the earliest on this CPU, but we run | |
1611 | * hrtimer_interrupt after we migrated everything to | |
1612 | * sort out already expired timers and reprogram the | |
1613 | * event device. | |
54cdfdb4 | 1614 | */ |
a6037b61 | 1615 | enqueue_hrtimer(timer, new_base); |
41e1022e | 1616 | |
b00c1a99 TG |
1617 | /* Clear the migration state bit */ |
1618 | timer->state &= ~HRTIMER_STATE_MIGRATE; | |
c0a31329 TG |
1619 | } |
1620 | } | |
1621 | ||
d5fd43c4 | 1622 | static void migrate_hrtimers(int scpu) |
c0a31329 | 1623 | { |
3c8aa39d | 1624 | struct hrtimer_cpu_base *old_base, *new_base; |
731a55ba | 1625 | int i; |
c0a31329 | 1626 | |
37810659 | 1627 | BUG_ON(cpu_online(scpu)); |
37810659 | 1628 | tick_cancel_sched_timer(scpu); |
731a55ba TG |
1629 | |
1630 | local_irq_disable(); | |
1631 | old_base = &per_cpu(hrtimer_bases, scpu); | |
1632 | new_base = &__get_cpu_var(hrtimer_bases); | |
d82f0b0f ON |
1633 | /* |
1634 | * The caller is globally serialized and nobody else | |
1635 | * takes two locks at once, deadlock is not possible. | |
1636 | */ | |
731a55ba | 1637 | spin_lock(&new_base->lock); |
8e60e05f | 1638 | spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING); |
c0a31329 | 1639 | |
3c8aa39d | 1640 | for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) { |
ca109491 | 1641 | migrate_hrtimer_list(&old_base->clock_base[i], |
37810659 | 1642 | &new_base->clock_base[i]); |
c0a31329 TG |
1643 | } |
1644 | ||
8e60e05f | 1645 | spin_unlock(&old_base->lock); |
731a55ba | 1646 | spin_unlock(&new_base->lock); |
37810659 | 1647 | |
731a55ba TG |
1648 | /* Check, if we got expired work to do */ |
1649 | __hrtimer_peek_ahead_timers(); | |
1650 | local_irq_enable(); | |
c0a31329 | 1651 | } |
37810659 | 1652 | |
c0a31329 TG |
1653 | #endif /* CONFIG_HOTPLUG_CPU */ |
1654 | ||
8c78f307 | 1655 | static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self, |
c0a31329 TG |
1656 | unsigned long action, void *hcpu) |
1657 | { | |
b2e3c0ad | 1658 | int scpu = (long)hcpu; |
c0a31329 TG |
1659 | |
1660 | switch (action) { | |
1661 | ||
1662 | case CPU_UP_PREPARE: | |
8bb78442 | 1663 | case CPU_UP_PREPARE_FROZEN: |
37810659 | 1664 | init_hrtimers_cpu(scpu); |
c0a31329 TG |
1665 | break; |
1666 | ||
1667 | #ifdef CONFIG_HOTPLUG_CPU | |
94df7de0 SD |
1668 | case CPU_DYING: |
1669 | case CPU_DYING_FROZEN: | |
1670 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu); | |
1671 | break; | |
c0a31329 | 1672 | case CPU_DEAD: |
8bb78442 | 1673 | case CPU_DEAD_FROZEN: |
b2e3c0ad | 1674 | { |
37810659 | 1675 | clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &scpu); |
d5fd43c4 | 1676 | migrate_hrtimers(scpu); |
c0a31329 | 1677 | break; |
b2e3c0ad | 1678 | } |
c0a31329 TG |
1679 | #endif |
1680 | ||
1681 | default: | |
1682 | break; | |
1683 | } | |
1684 | ||
1685 | return NOTIFY_OK; | |
1686 | } | |
1687 | ||
8c78f307 | 1688 | static struct notifier_block __cpuinitdata hrtimers_nb = { |
c0a31329 TG |
1689 | .notifier_call = hrtimer_cpu_notify, |
1690 | }; | |
1691 | ||
1692 | void __init hrtimers_init(void) | |
1693 | { | |
1694 | hrtimer_cpu_notify(&hrtimers_nb, (unsigned long)CPU_UP_PREPARE, | |
1695 | (void *)(long)smp_processor_id()); | |
1696 | register_cpu_notifier(&hrtimers_nb); | |
a6037b61 PZ |
1697 | #ifdef CONFIG_HIGH_RES_TIMERS |
1698 | open_softirq(HRTIMER_SOFTIRQ, run_hrtimer_softirq); | |
1699 | #endif | |
c0a31329 TG |
1700 | } |
1701 | ||
7bb67439 | 1702 | /** |
654c8e0b | 1703 | * schedule_hrtimeout_range - sleep until timeout |
7bb67439 | 1704 | * @expires: timeout value (ktime_t) |
654c8e0b | 1705 | * @delta: slack in expires timeout (ktime_t) |
7bb67439 AV |
1706 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL |
1707 | * | |
1708 | * Make the current task sleep until the given expiry time has | |
1709 | * elapsed. The routine will return immediately unless | |
1710 | * the current task state has been set (see set_current_state()). | |
1711 | * | |
654c8e0b AV |
1712 | * The @delta argument gives the kernel the freedom to schedule the |
1713 | * actual wakeup to a time that is both power and performance friendly. | |
1714 | * The kernel give the normal best effort behavior for "@expires+@delta", | |
1715 | * but may decide to fire the timer earlier, but no earlier than @expires. | |
1716 | * | |
7bb67439 AV |
1717 | * You can set the task state as follows - |
1718 | * | |
1719 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1720 | * pass before the routine returns. | |
1721 | * | |
1722 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1723 | * delivered to the current task. | |
1724 | * | |
1725 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1726 | * routine returns. | |
1727 | * | |
1728 | * Returns 0 when the timer has expired otherwise -EINTR | |
1729 | */ | |
654c8e0b | 1730 | int __sched schedule_hrtimeout_range(ktime_t *expires, unsigned long delta, |
7bb67439 AV |
1731 | const enum hrtimer_mode mode) |
1732 | { | |
1733 | struct hrtimer_sleeper t; | |
1734 | ||
1735 | /* | |
1736 | * Optimize when a zero timeout value is given. It does not | |
1737 | * matter whether this is an absolute or a relative time. | |
1738 | */ | |
1739 | if (expires && !expires->tv64) { | |
1740 | __set_current_state(TASK_RUNNING); | |
1741 | return 0; | |
1742 | } | |
1743 | ||
1744 | /* | |
1745 | * A NULL parameter means "inifinte" | |
1746 | */ | |
1747 | if (!expires) { | |
1748 | schedule(); | |
1749 | __set_current_state(TASK_RUNNING); | |
1750 | return -EINTR; | |
1751 | } | |
1752 | ||
1753 | hrtimer_init_on_stack(&t.timer, CLOCK_MONOTONIC, mode); | |
654c8e0b | 1754 | hrtimer_set_expires_range_ns(&t.timer, *expires, delta); |
7bb67439 AV |
1755 | |
1756 | hrtimer_init_sleeper(&t, current); | |
1757 | ||
cc584b21 | 1758 | hrtimer_start_expires(&t.timer, mode); |
7bb67439 AV |
1759 | if (!hrtimer_active(&t.timer)) |
1760 | t.task = NULL; | |
1761 | ||
1762 | if (likely(t.task)) | |
1763 | schedule(); | |
1764 | ||
1765 | hrtimer_cancel(&t.timer); | |
1766 | destroy_hrtimer_on_stack(&t.timer); | |
1767 | ||
1768 | __set_current_state(TASK_RUNNING); | |
1769 | ||
1770 | return !t.task ? 0 : -EINTR; | |
1771 | } | |
654c8e0b AV |
1772 | EXPORT_SYMBOL_GPL(schedule_hrtimeout_range); |
1773 | ||
1774 | /** | |
1775 | * schedule_hrtimeout - sleep until timeout | |
1776 | * @expires: timeout value (ktime_t) | |
1777 | * @mode: timer mode, HRTIMER_MODE_ABS or HRTIMER_MODE_REL | |
1778 | * | |
1779 | * Make the current task sleep until the given expiry time has | |
1780 | * elapsed. The routine will return immediately unless | |
1781 | * the current task state has been set (see set_current_state()). | |
1782 | * | |
1783 | * You can set the task state as follows - | |
1784 | * | |
1785 | * %TASK_UNINTERRUPTIBLE - at least @timeout time is guaranteed to | |
1786 | * pass before the routine returns. | |
1787 | * | |
1788 | * %TASK_INTERRUPTIBLE - the routine may return early if a signal is | |
1789 | * delivered to the current task. | |
1790 | * | |
1791 | * The current task state is guaranteed to be TASK_RUNNING when this | |
1792 | * routine returns. | |
1793 | * | |
1794 | * Returns 0 when the timer has expired otherwise -EINTR | |
1795 | */ | |
1796 | int __sched schedule_hrtimeout(ktime_t *expires, | |
1797 | const enum hrtimer_mode mode) | |
1798 | { | |
1799 | return schedule_hrtimeout_range(expires, 0, mode); | |
1800 | } | |
7bb67439 | 1801 | EXPORT_SYMBOL_GPL(schedule_hrtimeout); |